| Summary: | Giving the increasing penetration of intermittent wind power in the liberalized electricity market, wind power forecasting (WPF) is a topic of growing importance [Kariniotakis, 2017]. The number of papers on the field WPF evaluating the statistical performance has increased rapidly, while only a proportion of the former studies focus on the economic benefit of WPF. In this study we have answered how well a set of wind power forecasting (WPF) models works as day-ahead trading strategies for a 54MW wind power park. The performance evaluation is based on both statistic and economic measures. The wind power park is located in Northern Norway in a region with complex terrain and an arctic and coastal climate. The WPF models are applied on weather forecasts provided by two numerical weather prediction (NWP) models i.e., MEPS and AROME Arctic, operated by the Norwegian Meteorological Institute (MET Norway). When applied on MEPS forecasts of wind speed and wind direction, and the statistical performance measures are evaluated over a test period, it is evident that the multilayer perceptron (MLP) model provides the lowest NRMSE of 21.4%. Compared to a current forecasting method of the responsible power trader (ISHK model), the MLP model shows an improvement of 4.0%. Further enhancement of the accuracy of the MLP model is attained by adding air pressure as the third input feature. The resulting NRMSE is 20.9% of installed capacity. This corresponds to a 6.3% improvement compared to the ISHK model, which verify that the MLP model can compete with a current forecasting method of the responsible power trader on statistical measures. When it comes to the economic perspective, given a single-price system of the power market, the naive persistence model surprisingly shows the highest revenue for the power producer. A total revenue of 16.42 MNOK is obtained, where the imbalance revenue accounts for 200 kNOK. However, considering both statistical and economic measures it is evident that the ISHK model is the most effective ...
|
|---|